Repository navigation
✨ images: build and publish hive-mx (Betty C2 part 2) - #18428
Conversation
|
Hive status — updated in place; edits do not notify. Full history is in the Activity Monitor. @clubanderson this PR is held at Failing checks:
Next: the hive does not push to your branch. Push a new head (or reply here); CI and the review swarm re-run automatically and the PR returns to the normal path. Review findings from This PR edits workflow files ( What I read: Open question for the maintainer: this adds a new externally published package and a new required-looking hive/awaiting-ci: 'Sign-off survives the squash' fails on sign-off attribution (a commit's Signed-off-by must name its author; the hive cannot certify the human's DCO). Needs the PR author to re-sign (git rebase --signoff / commit --amend -s) and force-push. native-queue-policy is also red. Deferring this kick. 🐝 Hive Agent: — hive: agent=scanner backend=claude model=claude-sonnet-5-5 claude=2.1.325 |
70a43d2
ℹ️ Sentinel notice — informational; author is trusted (owner), merge is not blockedHive flagged this PR (author @clubanderson, head
Hive did not add the |
There was a problem hiding this comment.
Decision needed: @clubanderson should confirm the new build-mx/merge-mx jobs in .github/workflows/docker.yml (GHCR push with GITHUB_TOKEN, cross-org mirror login) are acceptable to publish a new public image.
Routing, not a defect finding: this PR edits .github/workflows/* and adds a new published image (external side effect), which I always send to a person.
Verified in the diff: all new uses: are SHA-pinned with version comments; docker/login-action uses secrets.GITHUB_TOKEN (docker.yml diff lines ~70, ~155) and a third login for the cross-org mirror (~197); no pull_request_target; the push gating copies build's needs.gate.outputs.push condition. I did not find a permissions widening in the added lines, but I did not confirm the job-level permissions: blocks match build-hub/merge-hub; a maintainer should check that. The body's description matches the file list.
— hive: agent=reviewer backend=claude model=claude-sonnet-5-5 claude=2.1.325
|
@clubanderson this PR is back in your court (
Once the new head lands, CI and the review swarm re-run automatically and the PR returns to the normal path. Current head: Updates (newest last):
|
There was a problem hiding this comment.
Decision needed: @clubanderson should decide whether .github/workflows/docker.yml may start publishing a new public image ghcr.io/<owner>/hive-mx (and mirror it cross-org) from every v5 push.
Routing to a human, not a defect: this PR edits .github/workflows/docker.yml (new build-mx/merge-mx jobs, +228 lines) and adds a new externally published image plus a cross-org mirror, which is a trust-surface and external-publish change.
What I read: build-mx/merge-mx (docker.yml diff) are a faithful copy of the hub jobs. They keep the same SHA-pinned actions, contents-less token use, the needs.gate.outputs.push gating, the same v5 true candidate publish args, and the same hard-fail when the cross-org secret is missing on hivecommons. src/Dockerfile.mx pins both base images by digest and runs as 65532. The four count assertions in test-publish-image-tags.sh and test_arm64_lane_probes_pr_code.sh were bumped 3 to 4 consistently with the new jobs. I found no logic error.
Note for the decider: releases.md documents that hive-mx is deliberately not retagged :v<version>, promoted to :stable or pruned, so it will accumulate untagged-by-release images until a follow-up lands. That seems intentional per the PR body.
— hive: agent=reviewer backend=claude model=claude-sonnet-5-5 claude=2.1.325
|
hive/scanner: 'Sign-off survives the squash' fails on d54fbdc and merge commit c981514: 🐝 Hive Agent: — hive: agent=scanner backend=claude model=claude-sonnet-5-5 claude=2.1.325 |
src/Dockerfile.mx: the digest-pinned golang:1.27-alpine builder (go.mod and the Crux go.mod first for a module-download layer) and a static hive-mx on gcr.io/distroless/static-debian12:nonroot (index digest pinned, uid 65532, ports 2525 and 8080). docker.yml gains build-mx and merge-mx, copies of the hub jobs: per-arch build by digest for amd64 and arm64, manifest list via publish-image-tags.sh and the cross-org mirror. v2-ci's pre-merge docker matrix builds it too (integration CI regenerated); the publish-tag and arm64-lane contract tests count four image rows; TRUSTED_SCRIPTS_TREE re-pinned. Part of #18391, #18395. Signed-off-by: Andy Anderson <andy@clubanderson.com>
70a43d2 to
dafbd9e
Compare
There was a problem hiding this comment.
Decision needed: @clubanderson should confirm the new TRUSTED_SCRIPTS_TREE value in .github/scripts/integration-governance.cjs is the intended regenerated tree hash, since this PR edits the workflow/CI trust surface and publishes a new GHCR image with the cross-org PAT.
Routing, not rejecting: I read the diff and found no defect, but it changes CI trust surfaces, which I don't approve alone.
What I read (diff at head dafbd9e):
build-mx/merge-mxin.github/workflows/docker.ymlmirrorbuild-hub/merge-hub: same action SHA pins, samegateconditions, samepublish-image-tags.shargs (v5 true candidate), and the same cross-org credential gate (secrets[needs.gate.outputs.cross_secret]). Nothing new in permissions or secrets handling.- The
mxmatrix rows inv2-ci.ymland the two generated integration workflows are consistent with each other. src/Dockerfile.mxuses digest-pinned bases and runs as 65532 with no shell.test-publish-image-tags.shandtest_arm64_lane_probes_pr_code.shcounts (3→4) match the fourpublish-image-tags.sh, run-number label,INCLUDE_LATESTand arm64-runner occurrences the new jobs add.
Open question for a human (not verified): integration-governance.cjs:12 swaps the hardcoded TRUSTED_SCRIPTS_TREE (b60380f… → ae5964e…). The comment above it says a caller-supplied value must never choose the verifier tree, so this literal is a trust anchor. I can't recompute the tree hash of the .github closure from here; someone should confirm it comes from generate-integration-ci.py as the body says and not a hand edit.
Also: the new image publishes :latest and :candidate for a service that receives inbound mail; releases.md says it is deliberately outside retag/promote, which reads as intentional.
— hive: agent=reviewer backend=claude model=claude-sonnet-5-5 claude=2.1.325
|
[APPROVALNOTIFIER] This PR is APPROVED Approval requirements bypassed by manually added approval. This pull-request has been approved by: The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
Part 2 of #18395 (Betty C2, email in; plan #18391). Follows #18423 (hive-mx itself).
What
src/Dockerfile.mx: the same digest-pinnedgolang:1.27-alpinebuilder assrc/Dockerfileandsrc/Dockerfile.hub(cross-compiled perTARGETARCH,CGO_ENABLED=0,-trimpath), and a runtime ofgcr.io/distroless/static-debian12:nonrootpinned to its multi-arch index digest: one static binary, CA certificates, no shell, uid/gid 65532, ports 2525 (SMTP; the Service maps 25 to it) and 8080 (health). Time zone data is linked into the binary (time/tzdatain cmd/hive-mx).docker.yml:build-mxandmerge-mx, copies ofbuild-hub/merge-hub: per-arch build by digest on the same runners (linux/amd64 and linux/arm64, hive-oke is arm64), manifest list and moving tags throughpublish-image-tags.sh(:<sha7>,:<branch>-latest,:latest,:candidateon v5) asghcr.io/<owner>/hive-mx, and the cross-org mirror.v2-ci.yml: the pre-mergedockermatrix buildssrc/Dockerfile.mxtoo (docker (mx)); the generated integration workflows are regenerated withgenerate-integration-ci.py.releases.md: hive-mx is published with the others but not yet retagged:v<version>, promoted to:stableor pruned (adding it to those loops would fail a release whose SHA predates the image); deploy it pinned by tag and digest.🤖 Generated with Claude Code